A NEW APPROACH FOR MODELING UNCERTAINTY IN REMOTE SENSING CHANGE DETECTION PROCESS

Abstract

Land use/cover change mapping is one of the basic tasks for environmental monitoring and management. Since the change maps are usually utilized in the planning and decisionmaking processes, therefore identification of the certainty and reliability of these maps is very important in many applications. Unfortunately in many studies only the pixel-based spectral and probabilistic measures as obtained from the classification approaches such as the maximum likelihood algorithm are used for uncertainty estimation. In this work, a new approach has been developed which is based on the pixel-based and object-based probability information as well as the spatial parameters including the distance, neighborhood, region size and the type of change. Two Landsat TM images of Isfahan urban area (Iran), acquired in 1990 and 1998 have been co-registered using the first order polynomial and the nearest neighbor resampling approaches .The registered images have been classified using the Maximum Likelihood Classification Algorithm (MLC) and the probabilistic measures have been generated. Using different spatial analysis functions, for modeling the change of agriculture to urban areas the relevant spatial parameters have been extracted. The probabilistic and spatial parameters have been integrated through the logistic regression modeling approach to model uncertainty of change of agriculture to urban areas. The Relative Operating Characteristics (ROC) index has been used for validation of the model and it has been estimated to be 0.9944, which is an indicator of the good model fitting. Further test and development of the proposed approach is under investigation.